>>12907826One interesting problem is how these images are formed in the first place. The sensors of the machine give you some sort of data based on the object they're imaging. So the problem is y = Ax + e, where y is the sensor data, x is the subject, e is an error term, and A is the discretised matrix projecting the object onto the sensors. The problem then becomes one of determining what x is, given that you know y, which you can solve with, for instance, truncated singular value decomposition or Landweber-Fridman iteration etc.
Another super interesting problem is taking a large data set of human genomes, using PCA to reduce the dimensionality and then clustering them with, say, K-means to figure out what kind of human races there might be.
